SRResNet Performance Enhancement Using Patch Inputs and Partial Convolution-Based Padding

Due to highly underdetermined nature of Single Image Super-Resolution (SISR) problem, deep learning neural networks are required be more deeper solve the problem effectively. One successful in (SR) is ResNet which can render capability with help skip connections. However, zero padding (ZP) scheme network restricts benefits connections SRResNet and its performance as ratio number pure input data that padded increases. In this paper. we consider Partial Convolution based Padding (PCP) instead ZP SR problem. Since training using patch images advantageous many aspects such image complexities, compared single full one. The experimental results show better than ones PCP one ZP.